Thomas Lumley • 6/26/2019

Denominator degrees of freedom in svyglm

This working note explores the complexities of determining denominator degrees of freedom for hypothesis tests in `svyglm` models from the R survey package. It discusses how the standard formula (design df minus number of parameters) is insufficient for stratified, clustered designs, especially when covariates exist at the cluster level. The author analyzes the rank of the variance-covariance matrix and proposes a more nuanced calculation that accounts for cluster-level predictors.

0 comments

#Statistical Modeling #Survey Analysis #Variance Estimation